I’ve come to think of the collection of tools I use as a kind of menagerie of models. I thought I’d share how I use them.
Full Show Notes: https://thejaymo.net/2024/08/24/2420-a-menagerie-of-models/
- Experience.Computer: https://experience.computer/
- Worldrunning.guide: https://worldrunning.guide/
- Subscriber Zine! https://startselectreset.com/
Permanently moved is a personal podcast 301 seconds in length, written and recorded by @thejaymo
Subscribe to the Podcast: https://permanentlymoved.online/
Quarterly zine; my gift to you ✉️
A Menagerie of AI Models
At a friend’s birthday drinks last week, the topic of AI came up. One person mentioned they tried ChatGPT last year but haven’t bothered with it since – saying they don’t see much use for it in their lives.
Which surprised me, because AI tools are everywhere in my life, and have been for a while. I’ve come to think of the collection of tools I use as a kind of menagerie of models.
And I thought I’d share how I use them.
First off, the covers for this show since episode 2129 have all been generated by VQGAN, the image synthesis tool I talked about playing with. Around that same time, I began using the AI-assisted video and podcast editor, Descript, to produce this show. I draft episodes in Google Docs, record the audio and video, use Descript’s transcription tool, and then edit the audio and video via a text pane.
If I didn’t have video to deal with, I think I could probably edit an audio-only version of this show in about five minutes at this point. Which is quicker than reading the script aloud.
I started paying for ChatGPT in 2022 within 20 minutes of first trying it. About a year ago, I began going for walks. Talking to myself, using hands-free in-ear headphones to record my thoughts with my Pixel’s onboard recorder app. Sometimes if the Thames Path is a bit busy I feel like a bit of a basket case – but I hope people just think I’m on the phone. Anyways, Pixel recorder transcribes speech in realtime and I get home with sometimes thousands of words of rough text. The transcription then syncs via Google Docs, and then I ask GPT-4 to tidy it up into full sentences.
Here’s my prompt:
Hi GPT-4! What follows is a poor quality transcription. Could you please do your best to interpret it and clean it up into full sentences? Keep as much of the original text as possible. Don’t condense, summarise or editorialise it.
This creates what I call the zero draft. Raw material that needs sculpting into the first draft. The first episode of 301 I made this way was the one about the new Zelda game in May of last year – Episode 2318. Sometimes, I’ll go for a walk, repeat the GPT process, and throw the output in my Notion for safekeeping. The Doc Web a post I published on my blog this week began life this way several weeks ago.
As I’ve become more comfortable with transcription and speech-to-text in general, the concept of speaking to my computer has opened up. I now use my M2 MacBook’s livetext tool all the time. I’ve even have it mapped it to my Globe function key. I’ll hit it and speak the rest of an email or paragraph I’m writing aloud. Despite having learned to put one word after another over the last 10 years, and making episodes about typing at the speed of thought, speech is just so much faster than fingers.
The liberal use of voice and transcription has transformed how I approach text production. If this technology had been around when I was a teenager, I think I wouldn’t have struggled so much with my school homework. Even now, approaching 40, this dyslexic man still finds it daunting to face a blank page. But with speech to text, as long as you have something to say, that hurdle is gone.
In recent months, I’ve been experimenting with a code writing app called Cursor, which is like GitHub’s Co-pilot but way better. During lockdown, I tried learning Python to play with my 3D depth camera but lost interest when things didn’t immediately go to plan . But Cursor has changed that. Though I still have no idea how to code – I’ve been running it. If the aim of learning to code is to run what you’ve written – then Cursor has got me from zero to one.
Recently, after installing Meta’s LLaMA 8B model locally on my laptop with oLLaMa. It’s integrated into my Spotlight replacer, Alfred. Having a GPT-3 level intelligence running on my machine using less RAM than a web browser is a revelation. It’s a glimpse of the future. I can copy and paste text, interrogate it via a screen overlay, and get the information I need without breaking flow.
The other day, I asked it to define the difference between denotative and connotative speech. It all happened seamlessly, without having to leave Notion. My computer told me exactly what I needed to know.
To recap: In the model menagerie I have, AI for image generation, transcription, podcast editing, and a local one for simple tasks. I also use Perplexity, the AI-powered search engine regularly. The ability to co-navigate search with an conversational interface is a game changer. It’s so much better than Google.
Within three years AI tools have deeply embedded in creative life. I don’t use them to generate slop or produce stuff from scratch – ever. But their abilities and usefulness to me is undeniable. They’ve sped up some parts of my creative processes and created entirely new ones that just weren’t possible before.
Which is why that conversation at my friend’s birthday was so surprising to me. AI might not be for everyone, but its impact on my life has been transformative, and they are only going to improve.
Subscribing to SSRZ supports my online work and creative projects.
As a thank you, I send you my zine four times a year, just like it’s 1994.
No spam. No email. Cancel at any time.
Leave a Reply